基于数据挖掘和分布式智能体的产品平台逆向建模与变异性分析

项目名称： 基于数据挖掘和分布式智能体的产品平台逆向建模与变异性分析

项目编号： No.51275362

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 机械、仪表工业

项目作者： 彭卫平

作者单位： 武汉大学

项目金额： 62万元

中文摘要： 企业存在大量的产品数据有待处理，现有的分析方法注重产品的相似性，易导致产品平台定义不完备等缺陷。本项目将智能体作为产品平台的基本模型单元，并引入主特征向量，通过插值排序和建立参数方程，揭示变异条件下产品结构和尺度的变化规律，探索产品建模新理论。面向模型相似性和差异性分析，研究基于贝叶斯权值的多层前馈神经网络分类和加权模糊聚类、基于主特征向量筛选的结构匹配与剪枝、基于凝聚算法的多层次聚类、基于Apriori算法的多空间关联并行挖掘等算法和工具，形成产品数据分析新方法。运用这些理论和方法，对给定实例产品数据库，实现产品模型、结构同构类与非同构聚类、功能等效类与非等效聚类、工艺等价类与非等价聚类、及其包含性结构的提取和输出，构造一个分布式产品平台。通过HLA/RTI仿真技术，研究产品平台的行为特征和演化规律，以验证理论和方法的有效性。为构建面向大批量定制设计的产品平台提供新的理论、方法和工具。

中文关键词： 产品数据挖掘；变异性分析；逆向建模；参数化模型；产品平台

英文摘要： There is a lot of product data need to be processed in many enterprises. Because of focus only on product similarity, the existing methods of product data analysis often led to incomplete definition of product platform. In order to solve this problem, a new theory of product modelling is presented in this research. Based on this new theory, agent with main feature attribute vector is used as basic model unit of a product platform. By interpolating, sorting and establishing parameter equation for the main feature attribute vector which show internal state of agents, the change rule and variation law of product structure and its scale are determined. Besides, many new methods for the similarity and difference analysis of product models are also presented. These methods include (1) multilayer feedforward neural network classification based Bayes weighting; (2) weighted fuzzy clustering with c-mean; (3) structure matching and pruning selectively based on main feature attribute vector; (4) multilayer clustering based on condensing algorithm; (5) parallel mining of multispace incidence relation based on Apriori algorithm, etc. According to above theory and methods, product models are extracted from a given case product data base. And based on analysis of the product models, the isostructure classes and non-isostructur

英文关键词： product dada mining；veriability analysis；reverse modelling；parameterized model；product platform